Chinese web giant Tencent says new US GPU ban ‘sets everyone back in China’
First reported by The Register ·
Access to advanced AI training hardware is now more restricted and costly.
Tencent president Martin Lau stated that new US restrictions on exporting advanced GPUs to China will temporarily hinder the company's ability to train AI models. He explained that access to these high-end GPUs is crucial for Tencent's large-scale AI development, including its agentic assistant WorkBuddy, which processes over 100 trillion tokens daily. Lau noted that while inferencing workloads might be less affected due to the availability of domestic chips, training efficiency will decrease in the short term. He also anticipates that the ban will accelerate China's development of domestic GPU supply chains and alternative computing solutions, as Chinese companies will increasingly rely on local alternatives for both training and inferencing.
The US ban on advanced GPU exports to China, including those from Nvidia, directly impacts Tencent's ability to train its large language models efficiently. This forces the company to seek alternative, potentially less performant or more expensive, solutions for its AI development efforts. The disruption highlights the critical dependence of major AI players on a limited number of hardware suppliers and the geopolitical implications of such dependencies.
While Tencent anticipates leveraging domestic Chinese chip manufacturers for inferencing tasks, the primary impact is on model training. This restriction is expected to accelerate China's drive for self-sufficiency in AI hardware, potentially leading to increased investment and innovation in domestic GPU technology. The long-term consequence could be a bifurcated global AI hardware market, with China becoming more isolated but also more robust in its indigenous capabilities.
AI-written summary. May contain errors.